Share
Embed
Empirically Assessing Opportunities for Prefetching and Caching in Mobile Apps
Presentation slides of the paper "Empirically Assessing Opportunities for Prefetching and Caching in Mobile Apps" at the International Conference on Automated Software Engineering (ASE) 2018.
The presentation can be found at https://youtu.be/iT1pNy370ZM
Yixue Zhao, PhD
September 06, 2018
Transcript
81 slides
-
Empirically Assessing Opportunities for Prefetching and Caching in Mobile Apps
Yixue Zhao, Paul Wat, Marcelo Schmitt Laser, and Nenad Medvidović University of Southern California September 6, 2018 @Montpellier -
Prefetching & Caching immediate response local cache ICSE 2018. Leveraging
Program Analysis to Reduce User-Perceived Latency in Mobile Applications. -
Browser Mobile App Browser vs. Mobile App Ø Lack of
work Ø Large body of work ? -
Browser Mobile App Browser vs. Mobile App Ø Lack of
work Ø “Light” requests Ø Large body of work Ø “Rich” requests ? -
Browser Mobile App Browser vs. Mobile App Ø Lack of
work Ø “Light” requests Ø Single round-trip time Ø Large body of work Ø “Rich” requests Ø Loading multiple sub-resources ? -
Browser Mobile App Browser vs. Mobile App Ø Lack of
work Ø “Light” requests Ø Single round-trip time Ø Prefetch separate requests Ø Large body of work Ø “Rich” requests Ø Loading multiple sub-resources Ø Prefetch sub-resources ? -
Conclusion Ø Browser à Mobile app Ø Lack of prefetching
& caching in mobile apps Browser vs. Mobile App -
Research Questions Ø Prefetchability of HTTP requests Ø Cacheability of
HTTP responses Ø Redundancy of HTTP requests -
GET requests Ø Read-only Ø No side effects Ø Safe
to prefetch RQ: Prefetchability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests -
1. What is the number of GET requests per app?
2. What is the percentage of GET requests among all HTTP requests? 3. How prevalent are GET requests across different app categories? RQ: Prefetchability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests -
RQ: Cacheability Ø Prefetchability of HTTP requests Ø Cacheability of
HTTP responses Ø Redundancy of HTTP requests -
Prefetchable ≠ Cacheable Ø Cache staleness Ø Rely on Expires,
Cache-Control Ø Trustworthy? RQ: Cacheability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests -
4. How prevalent are Expires headers? 5. Are Expires headers
trustworthy? 6. How prevalent are Cache-Control headers? 7. Are Cache-Control headers trustworthy? RQ: Cacheability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests -
RQ: Redundancy Ø Prefetchability of HTTP requests Ø Cacheability of
HTTP responses Ø Redundancy of HTTP requests -
Redundant requests Ø Ostensibly redundant: same URL Ø Truly redundant:
same URL & same response RQ: Redundancy Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests Cache me J -
8. How prevalent are redundant HTTP requests? 9. Are the
identified ostensibly redundant requests truly redundant? RQ: Redundancy Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests -
Workflow Overview Send Requests t=0, 10, 30, 60s #req ≥
4 Instrumented Apps App Testing Monkey Exerciser download Manual Examination Initial Apps -
Ø 1,687 top-ranked apps Ø 33 app categories e.g., Beauty,
Tools Workflow: Download download Google Play Initial Apps 1,687 -
Ø Instrumented 1,308 apps to capture needed information by Soot
Ø App size: 16 KB to 103.4 MB Ø #Request per app: 0 to 1,243 Workflow: Instrumentation download Google Play Initial Apps Instrumented Apps 1,308 1,687 -
Ø 3,000 random events by Monkey Ø NoxPlayer emulator Ø
Wi-Fi network settings Workflow: Testing download Google Play Initial Apps Instrumented Apps App Testing Monkey Exerciser 1,687 1,308 -
Workflow: Inspection Ø Low network usage Ø Manual inspection Instrumented
Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps -
Ø Low network usage Ø Manual inspection Ø 6 recurring
reasons e.g. obfuscation Workflow: Inspection Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps -
1 ≤ #Requests ≤ 3 Ø Load config files Ø
Log in with Facebook Ø Monitoring services Workflow: Inspection Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps -
#req ≥ 4 Workflow: Final Apps Final apps (#req ≥
4) Ø 451 apps Ø 33 app categories e.g. Beauty, Tools Ø #Req per app Avg: 35 Max: 1,243 -
Workflow Overview Send Requests t=0, 10, 30, 60s #req ≥
4 Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps -
Findings Ø Prefetchability of HTTP requests Ø Cacheability of HTTP
responses Ø Redundancy of HTTP requests -
Prefetchability of HTTP requests Ø What is the number of
GET requests per app? Min 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 -
Prefetchability of HTTP requests Ø What is the number of
GET requests per app? Max 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 -
Prefetchability of HTTP requests Ø What is the number of
GET requests per app? Avg 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 -
Prefetchability of HTTP requests Ø What is the number of
GET requests per app? 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 1,243 -
Prefetchability of HTTP requests Ø What is the number of
GET requests per app? 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 1,243 263 182 802 286 174 166 -
Prefetchability of HTTP requests Ø What is the percentage of
GET requests per app? 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 -
Prefetchability of HTTP requests Ø GET is pervasive à prefetching
opportunity 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 -
Prefetchability of HTTP requests Ø GET is pervasive à prefetching
opportunity Ø Highest avg: Beauty 94% Ø Lowest avg: Dating 43% 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 -
Prefetchability of HTTP requests 0 25 50 75 100 1
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Beauty 94% -
Prefetchability of HTTP requests 0 25 50 75 100 1
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Dating 43% -
Findings Ø Prefetchability of HTTP requests Ø Cacheability of HTTP
responses Ø Redundancy of HTTP requests -
Cacheability of HTTP responses Ø Expires & Cache-Control response headers
Ø Numbers Ø Percentages 0 5 10 15 20 25 30 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Percentages of Expires Numbers of Expires -
Cacheability of HTTP responses Avg of Expires percentages Avg of
Cache-Control percentages 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 -
Cacheability of HTTP responses Ø Avg: Expires 53%, Cache-Control 65%
Avg of Cache-Control percentages 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Avg of Expires percentages -
Cacheability of HTTP responses Ø Avg: Expires 53%, Cache-Control 65%
Ø Both headers are not always included Avg of Cache-Control percentages 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Avg of Expires percentages -
Cacheability of HTTP responses Instrumented Apps App Testing Manual Examination
Monkey Exerciser Send Requests t=0, 10, 30, 60s #req ≥ 4 download Google Play Initial Apps -
Cacheability of HTTP responses Ø Send original request @ time(t)
Ø Send same request @ time(t+x), x=10, 30, 60s t = original time x = time period after t -
Cacheability of HTTP responses Ø Untrustworthy scenarios 1. time(exp) ≤
time(t) 2. time(t) < time(exp) ≤ time(t+x) ∧ response@(t) = response@(t+x) 3. time(exp) > time(t+x) ∧ response@(t) ≠ response@(t+x) t = original time x = time period after t -
Cacheability of HTTP responses Ø Untrustworthy scenarios t = original
time x = time period after t t t + x request@t 2018 Sep 1 2018 Sep 6 -
Cacheability of HTTP responses Ø Untrustworthy scenarios t = original
time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 -
Cacheability of HTTP responses Ø Untrustworthy scenarios t = original
time x = time period after t t t + x expire time 2018 Sep 1 2018 Sep 6 -
Cacheability of HTTP responses Ø 1. expire before request is
sent t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 expire time: 1999 Jan 1 -
Cacheability of HTTP responses Ø 2. should expire but response
is unchanged t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 expire time: 2018 Sep 3 -
Cacheability of HTTP responses Ø 3. should not expire but
response is changed t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 expire time: 2018 Sep 10 -
Cacheability of HTTP responses Ø Avg trustworthy: Expires 25%, Cache-Control
77% (3x) Avg of trusted Expires Avg of trusted Cache-Control 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 -
Cacheability of HTTP responses Ø Avg trustworthy: Expires 25%, Cache-Control
77% (3x) Ø 0% and 100% apps Avg of trusted Expires Avg of trusted Cache-Control 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 -
Cacheability of HTTP responses Ø Should not depend on the
headers completely -
Cacheability of HTTP responses Ø Should not depend on the
headers completely Ø No reliable alternatives -
Cacheability of HTTP responses Ø Should not depend on the
headers completely Ø No reliable alternatives Ø Research opportunities J -
Findings Ø Prefetchability of HTTP requests Ø Cacheability of HTTP
responses Ø Redundancy of HTTP requests -
Redundancy of HTTP requests Ø Ostensibly redundant requests Ø Avg:
20% 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Percentages of ostensibly redundant requests -
Redundancy of HTTP requests Ø Truly redundant requests Ø Avg:
92% 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Percentages of truly redundant requests -
Redundancy of HTTP requests Ø Most redundant requests are truly
redundant Ø What does this tell us? -
Implications for prefetching Ø “Avg” app Ø #GET request 28
Ø Response time: 800ms Time saved: 22s -
Implications for caching Ø “Avg” app Ø #GET request 28
Ø Redundant requests: 20% Ø Truly redundant requests: 92% Ø #Cacheable request: 6 Time saved: 4s -
Summary Empirical study to understand HTTP characteristics in mobile apps
Ø Prefetchability of requests Ø Cacheability of responses Ø Redundancy of requests -
Lessons learned Ø Opportunity for prefetching & caching Ø App-specific
characteristics Ø Domain-specific characteristics Summary -
Summary Contributions Ø Bridge the gap between browser and mobile
app domain Ø Guidelines for developers Ø Motivate future research -
Behind the Scene: co-authors Marcelo Schmitt Laser Paul Wat advisor:
Nenad Medvidović -
Behind the Scene: co-authors + sponsors Marcelo Schmitt Laser Paul
Wat advisor: Nenad Medvidović